Raw hazelnut volatiles: challenges in defining odorant patterns related to sensory defects by comprehensive two-dimensional gas chromatography coupled with time-of-flight mass spectrometry

The study of the odor perception of a food, here raw hazelnut, poses severe challenges due to the high chemical dimensionality expressed in these mixtures of volatiles. In this scenario, the separation power and resolution enhancement, the improved sensitivity obtained by the effective band focusing in-space and the generation of structured separation patterns, are key-features that make comprehensive two-dimensional gas chromatography (GC×GC) with thermal modulators a platform of choice to achieve accurate and reliable fingerprinting results. Raw hazelnuts connoted by different sensory defects may show informative 2D patterns of volatiles that could be diagnostic for quality assessment. Volatiles from hazelnuts of different geographical origin and cultivar, selected by flash-profile descriptive analysis for the presence/or not of sensory defects, are sampled by headspace solid phase micro extraction (HS-SPME) with a DVB/CAR/PDMS df 50/30 μm 2 cm length fiber and analyzed by GC×GC-TOF MS adopting a polar × medium polarity column set-up. Their 2D patterns are then processed by combined Untargeted and Targeted fingerprinting (UT fingerprinting) and by visual features fingerprinting to highlight compositional differences. Finally, unsupervised and supervised chemometrics is adopted on 2D peaks quantitative descriptors to find reliable and informative peaks and peak-patterns for successful discrimination and classification of samples. The 2D patterns of volatiles from good quality and defected hazelnuts show a great complexity: 120 known compounds on about 350 2D peak-regions detectable, are reliably targeted (i.e.). UT fingerprinting delineates diagnostic patterns that clearly cluster samples with mouldy notes, rich in both saturated and unsaturated aldehydes, short chain fatty acids, linear alcohols and furanones, but fails with rancid samples due to the concurrent presence of additional perceptions like stale and solvent-like odors. To improve the fingerprinting effectiveness and to minimize the impact of confounding variables, a “model peak-pattern” is created after re-alignment and summation of 2D-contour plots from samples showing specific odor defects. The resulting cumulative image is then adopted as diagnostic probe for effective fingerprinting through visual features. Supervised chemometrics effectively extract informative analytes (known and unknown) with high discrimination role. In conclusion, the univocal identification of chemical patterns related to sensory defects confirms the effectiveness of high-informative fingerprinting by GC×GC-TOF MS and pattern recognition approaches based on template matching for sensory quality assessment of raw hazelnuts. On the other hand, visual features fingerprinting offers a unique option to minimize the effect of confounding variables, leading to more conclusive results.